|
Comments addressed by 8/23 release:
- Parser chokes on table within table formatting of this website: gets lost and misses first real table in CalculatorExample. -- WardCunningham . . . Test tables are now distinguished from other tables by enclosing them within <wiki> ... </wiki> tags.
Comments addressed by 8/27 release:
- You need to test for ExpectedErrors. Don't make the mistake made in ThePragmaticProgrammer. -- BretPettichord . . . The keyword error now inverts the error handling logic making exceptions good and results bad.
- I agree with the other reviewer to annotate the test results with a textual description along with the color encoding. I also noticed that the summary showed the actual result first and the expected result second. My suggestion would be to stick with JUnit's output sequence: expected ... but was ... -- FrankWestphal . . . I will modify the framework to label the values when more than one appear together in one cell. It is my intention that expected preceed actual results. -- WardCunningham
Comments addressed by 9/10 release:
- The download tar file contains Java source code which doesn't specify a package, hence it is using Java's default package. I have no problem with that, accept that when I import this code into Eclipse, it puts the source files under a package called Release.Source, which means it doesn't compile since the source code doesn't specify a package called Release.Source. Because this is destined to become public software, I'd suggest you use some package name instead of relying on Java's default package -- doing so will confuse fewer people and increase usability. -- JoshuaKerievsky . . . We've chosen the package name of "fit" for the framework and "eg" for the examples. -- WardCunningham
- On one of these projects, I didn't have a suite of AcceptanceTests yet, so that writing fixtures was harder. On this particular project, I kept writing tests that would first inject some interesting state into the system and exercise it in a later step. One thing that I missed in your framework for these kinds of test scenarios was something like a DontCare marker for the expected result column. -- FrankWestphal . . . I've added n/a as a distinguished value much like error. It stands for "not applicable" and means that the computation won't even be run. -- WardCunningham [note: on the subsequent release n/a was abandoned in favor of blank meaning "don't care".]
|